A Surrogate-Assisted Improved Many-Objective Evolutionary Algorithm

摘要

The many-objective evolutionary algorithm is an effective method to tackle many-objective optimization problems. We improve the two-archive2 algorithm (Two_Arch2) by adopting the Levy distribution and opposition-based learning strategy. In addition, we propose a hybrid adaptive strategy of the surrogate models. The criterion for evaluating the model quality is developed. In model management, selection of individuals is based on the criterion named angle penalized distance (APD). In the experiments, we make comparisons of the IGD among our algorithm and the other algorithms on the DTLZ and MaF test suites, which exhibits the superiority of the improved algorithm.